Claude Code for teams
Claude Code, set up so your team can trust it.
Out of the box, Claude Code does not know your project: its rules, its commands, the files it must never touch. I set it up on your codebase so it follows your rules, uses the stronger models only where they are needed, and stops before the risky actions you name. Then I hand it over with a written guide and a working session, so your team can run it without me.

Arrit Gashi
Independent consultant · Huntersville, North Carolina
What you get
Project rules the model follows
A CLAUDE.md written for your codebase: its structure, its commands, what done means, and the rules that must never be broken.
Guardrails enforced by code, not by the prompt
Permissions and hooks that refuse a write outside the project, block a risky command, or run your checks before a change is accepted.
Agents for the jobs you repeat
Subagents for review, research, testing and security checks, adapted from the ones I run every day and written for your stack.
The right model for each job
Each agent is set to the cheapest model that can do its work, with a rule for when a hard problem moves up to a stronger one.
A clear view of usage
A usage report that shows which model did the work and how many tokens each one used, so you can see which work uses the most.
A team that knows how to use it
A working session on your own repository, and a short written guide to every part, so the setup does not depend on me.
What I run myself
Nothing here is a demo. I use Claude Code every day as my main development environment, and 8 of the projects on this site are in use today.
21
agents in my own daily setup
Among them a director, a manager that gates every merge, a developer, a designer, a tester, security, legal, finance, player support and a cost monitor. Twelve run on Opus, seven on Sonnet, one on Haiku and one on the top model.
10,000+
players a month on the game they run
Pishpirik is a live multiplayer card game. A team of these agents builds it, tests it, checks its security and watches its costs.
216
deny rules around one project
One project reads its source material from 54 neighbouring folders, and its file-editing tools are refused in every one of them.
57
agents on one security audit
Before a launch: 46 raw findings, 43 verified, 3 refuted, 23 distinct vulnerabilities, 2 of them critical. The verdict was: do not ship.
Open source
model routing for Claude Code
An earlier version of the routing setup I use is public: agents pinned by task, an escalation ladder, and a skill that charts tokens per model.
3
Anthropic courses completed
- Building with the Claude API (2026)
- Claude Code 101 (2026)
- Claude Code in Action (2026)
Packages
Each one is a fixed scope at a fixed price, agreed in writing after the first call.
Setup
One repository that needs safe defaults and a few agents.
- CLAUDE.md with your project rules and commands
- Permissions and guard hooks
- Two or three agents for review, research or testing
- Tested on your repository
- A short guide to each part
Team rollout
A team that uses Claude Code every day and wants to control cost.
- Everything in Setup
- An agent team adapted to how you work
- Model routing and an escalation rule
- A usage report per model
- A working session with your team
Custom build
Claude built into your own product or internal tools.
- A feature built on the Claude API
- Or an agent workflow for one of your business processes
- Guards for the cases where the model is wrong
- Tests you can run again after a change
- Source and documentation handed over
How it runs
- 01
A 20-minute call
You tell me what your team builds and what slows it down. You get back a written scope and a fixed price. No charge for this part.
- 02
You share the repository
A copy is enough. I never ask for your passwords or production keys. Once handed over, everything runs on your own machines and accounts.
- 03
I build and test it on your code
The rules, hooks and agents are written for your project and run against it before you see them.
- 04
Handover and a working session
You get the files, a guide to each part, and a session with your team on your own repository.
Book a call
Email is the fastest route. Tell me what your team builds, how many people would use Claude Code, and what slows you down today. A paragraph is plenty.
Based in Huntersville, North Carolina. Remote work across US time zones, and European mornings. I am an independent consultant and am not affiliated with or endorsed by Anthropic. Claude and Claude Code are Anthropic products.